Messages

1

General / Re: Warped DSM in coastal strips (Phantom 4 RTK with GCPs)

« on: October 31, 2019, 06:51:33 PM »

Wow that is a fantastic paper, thanks for sharing this! Really describes the issue I have been seeing. I'll try out the off-nadir approach next time.

I used my own Trimble base station on a benchmark and did PPK using RTKLIB. I can get accuracy from the output of that process.

2

General / Re: Warped DSM in coastal strips (Phantom 4 RTK with GCPs)

« on: October 31, 2019, 05:59:36 PM »

This was pure Nadir. It is a pretty flat area and I was flying 400' AGL so this should be a fine setup.
I found that I needed to change my camera accuracy to a much smaller value. It was left at 10 m originally, but when changed to 0.1 m the model was significantly less warped. After this change, the 20 control points had a total error of 0.1 m.
I still see the same warp pattern when I subtract this DSM from the plane-based DSM, but now they generally agree to within 0.3 m, which is as much as I was hoping for. Maybe the warp is actually coming from the "control DSM" made from the plane?

3

General / Warped DSM in coastal strips (Phantom 4 RTK with GCPs)

« on: September 04, 2019, 01:43:55 AM »

I do coastal surveys and would like some advice for how to deal with surveys over long, thin strips near water.

I recently surveyed a coastal area of 6000 x 200 m with 2300 photos using the Phantom 4 RTK, and used 12 ~evenly spread GCPs with a GNSS system.
I compared the resulting DSM to two previous ones collected by more advanced systems, and found that my model has a clear warp of +/- 2 ft with a radial pattern. Without GCPs, the model warped up to 8 ft, so they certainly helped but did not fix the issue. See the attached image.

I performed the same survey style (P4RTK, GCPs, same height and overlap) on another location that was a 2500 x 1500 m area and the DSM matched perfectly with previous models. Even without GCPs the DSM did not have any warp, just needed a vertical shift. Thus, I think the warp I see may be due to processing a thin strip of land (and possibly the water nearby did not help).

I am looking for any insight into why this warp occurs, and suggestions to help process thin strips of land near water. I'm ready to test different solutions and report back. I am thinking that working in chunks may help, but I have not tested this yet.

The processing report is attached. My general workflow was add PPK cameras, Align, add GCPs, gradual selection and optimize cameras, dense cloud, DSM.

4

General / Estimated camera heights much lower than actual height

« on: August 18, 2018, 01:09:55 AM »

I flew three UAV surveys in the same manner as I always do. For two of them, after aligning images, the estimated camera heights are around 20 m AGL (also ~20 m NAVD88), when they should be around 120 m. The third survey was fine, camera estimates were within 5 m of 120 which is expected.

I tried running it raw (WGS coordinates with awful DJI-recorded height) and had this issue. The issue persisted even when correcting the flight heights: I converted coordinates to UTM. I converted heights to m NAVD88 by first taking offset between photo taken on ground right before launch and first survey photo taken. Offset is added to height of survey-grade GPS point taken at launch point. The issue also persisted when I placed ground control, unchecked camera coordinates, and ran alignment using referenced and also not referenced pre-selection.

The resulting sparse cloud is scaled correctly relative to itself, but the unplaced ground control are below it. Makes it hard to place them at first, but after placing one point on two images it guesses almost perfectly on the remaining images. When I placed ~30 ground control points the estimated camera height rose to 50-70 m AGL, still very low, but the ground control fit much better. The ground control z errors are 2-4 meters, which is unusually bad for this survey style and it's clear that the DSM will be warped by this amount. For the third survey which did not have this height problem, the marker total error is below 1 m which is expected.



I've been using this setup for many surveys and never had an issue until now.
Survey setup:
Phantom 3 Advanced - off the shelf setup
Camera GPS - Phantom 3 Advanced GPS - heights adjusted to NAVD88
Ground control - RTK-GNSS
DJI GS Pro app - flight heights held to 120 m AGL
Sidelap: 70%
Frontlap: 80%
3 flights done in succession over about 1 hour time period

Software setup (Tried alignments using these different settings and got the same error regardless):
Agisoft Photoscan 1.2.4 build 2399
~1000 photos
Alignment: high and highest quality
Referenced and non-referenced resulted in the same error (also tried unchecking camera coordinates, placed ~30 GCPs evenly spread)
Key point limits: tried 8000, 40000, 80000, and 0
Tie point limits: tried 40000, 400000, and 0
Mask: about 10% of image from edges, and no mask
Focal length: 3.61 mm
Pixel size: 0.00156192 mm
Frame camera

Any idea what could be causing this estimated height offset of 100 m, and a way to fix it?